L'ho trovato una volta quando cercavo un'abilità sfocata per i documenti quando la classe ConvolveOp non la tagliava (per le ragioni per cui si sta eseguendo). Fa una sfocatura gaussiana che è la sfocatura più naturale imho ... Speriamo che ti possa aiutare. Ho recuperato da questa pagina web: Java Image Processing ...
/*
** Copyright 2005 Huxtable.com. All rights reserved.
*/
package com.jhlabs.image;
import java.awt.image.*;
/**
* A filter which applies Gaussian blur to an image. This is a subclass of ConvolveFilter
* which simply creates a kernel with a Gaussian distribution for blurring.
* @author Jerry Huxtable
*/
public class GaussianFilter extends ConvolveFilter {
static final long serialVersionUID = 5377089073023183684L;
protected float radius;
protected Kernel kernel;
/**
* Construct a Gaussian filter
*/
public GaussianFilter() {
this(2);
}
/**
* Construct a Gaussian filter
* @param radius blur radius in pixels
*/
public GaussianFilter(float radius) {
setRadius(radius);
}
/**
* Set the radius of the kernel, and hence the amount of blur. The bigger the radius, the longer this filter will take.
* @param radius the radius of the blur in pixels.
*/
public void setRadius(float radius) {
this.radius = radius;
kernel = makeKernel(radius);
}
/**
* Get the radius of the kernel.
* @return the radius
*/
public float getRadius() {
return radius;
}
public BufferedImage filter(BufferedImage src, BufferedImage dst) {
int width = src.getWidth();
int height = src.getHeight();
if (dst == null)
dst = createCompatibleDestImage(src, null);
int[] inPixels = new int[width*height];
int[] outPixels = new int[width*height];
src.getRGB(0, 0, width, height, inPixels, 0, width);
convolveAndTranspose(kernel, inPixels, outPixels, width, height, alpha, CLAMP_EDGES);
convolveAndTranspose(kernel, outPixels, inPixels, height, width, alpha, CLAMP_EDGES);
dst.setRGB(0, 0, width, height, inPixels, 0, width);
return dst;
}
public static void convolveAndTranspose(Kernel kernel, int[] inPixels, int[] outPixels, int width, int height, boolean alpha, int edgeAction) {
float[] matrix = kernel.getKernelData(null);
int cols = kernel.getWidth();
int cols2 = cols/2;
for (int y = 0; y < height; y++) {
int index = y;
int ioffset = y*width;
for (int x = 0; x < width; x++) {
float r = 0, g = 0, b = 0, a = 0;
int moffset = cols2;
for (int col = -cols2; col <= cols2; col++) {
float f = matrix[moffset+col];
if (f != 0) {
int ix = x+col;
if (ix < 0) {
if (edgeAction == CLAMP_EDGES)
ix = 0;
else if (edgeAction == WRAP_EDGES)
ix = (x+width) % width;
} else if (ix >= width) {
if (edgeAction == CLAMP_EDGES)
ix = width-1;
else if (edgeAction == WRAP_EDGES)
ix = (x+width) % width;
}
int rgb = inPixels[ioffset+ix];
a += f * ((rgb >> 24) & 0xff);
r += f * ((rgb >> 16) & 0xff);
g += f * ((rgb >> 8) & 0xff);
b += f * (rgb & 0xff);
}
}
int ia = alpha ? PixelUtils.clamp((int)(a+0.5)) : 0xff;
int ir = PixelUtils.clamp((int)(r+0.5));
int ig = PixelUtils.clamp((int)(g+0.5));
int ib = PixelUtils.clamp((int)(b+0.5));
outPixels[index] = (ia << 24) | (ir << 16) | (ig << 8) | ib;
index += height;
}
}
}
/**
* Make a Gaussian blur kernel.
*/
public static Kernel makeKernel(float radius) {
int r = (int)Math.ceil(radius);
int rows = r*2+1;
float[] matrix = new float[rows];
float sigma = radius/3;
float sigma22 = 2*sigma*sigma;
float sigmaPi2 = 2*ImageMath.PI*sigma;
float sqrtSigmaPi2 = (float)Math.sqrt(sigmaPi2);
float radius2 = radius*radius;
float total = 0;
int index = 0;
for (int row = -r; row <= r; row++) {
float distance = row*row;
if (distance > radius2)
matrix[index] = 0;
else
matrix[index] = (float)Math.exp(-(distance)/sigma22)/sqrtSigmaPi2;
total += matrix[index];
index++;
}
for (int i = 0; i < rows; i++)
matrix[i] /= total;
return new Kernel(rows, 1, matrix);
}
public String toString() {
return "Blur/Gaussian Blur...";
}
}
fonte
2015-11-02 21:45:00
Cosa succede se si sostituisce 'int size = radius * 2 + 1;' con valori effettivi e vedere se la modifica di questo valore cambia la casella sfocata? – npinti
Vedere la sezione 'Lavorare in the Edge' in [questo articolo] (http://www.informit.com/articles/article.aspx?p=1013851&seqNum=5) per una soluzione alternativa. È necessario aumentare la dimensione dell'immagine rispetto alle dimensioni del kernel. – halex